conradlocke

/

krea2-identity-edit

like 715

I just tried v2. You have done an amazing job. Thank you very much.

Turbo model usage example with some findings :
Tried with a close up photo of a face with the settings below :
grounding pixels : 960
ref boost : 1.75 (4 was locking down the person's face in place)
step count : 7 ( higher values can cause issues in the eyes, lower values pushes the image faster for more preservation with less imagination aka identity preservation)
lcm, simple
worked great.

extra info: for stubborn images ref: boost 1.70 ,steps : 6

Thanks for this post, I don't find the workflow for this lora, Does anyone have it?

Thank you, really glad it is working well for you. And thanks for posting your settings, they are useful. Your observation is spot on: ref_boost 4 maximizes likeness but it locks the face in place, so for close-up work where you want a little more life, dialing it down to around 1.75 like you did is exactly right. It is meant to be a dial for that reason. Appreciate the detailed findings.

Thanks for this post, I don't find the workflow for this lora, Does anyone have it?

@vini76170 - It's easy to miss the workflow because it's not in this model page. This Lora requires a custom node called: Comfyui-Krea2Edit

Go here and you'll find the workflow and you'll need to install and get the custom node working.

https://github.com/lbouaraba/comfyui-krea2edit/tree/main/workflows

This Youtube video may help, but I don't believe it uses the updated workflow that I just linked to. Be sure to read ALL the notes in the workflow and in this discussion thread for good results.

https://www.youtube.com/watch?v=M_NWStvz9YQ